Criteria for considering studies for this review
Types of studies: The review is registered at PROSPERO with a registration number of CRD42020218517. We will follow the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (26) guidelines and include experimental (randomized trials (individually or cluster) and quasi-randomized trials) and observational studies with a concurrent comparison group (cohort (prospective and retrospective), controlled before-after studies and nested case-control studies). We will exclude case reports, case series, opinions, editorials, commentaries, letters, conference abstracts, and reviews or systematic reviews.
Types of participants: We will only include studies that enroll infants, living in low-, middle- or high-income countries. We will exclude animal studies and studies enrolling only infants with an existing illness (cancer, diabetes, metabolic disorder, HIV, congenital abnormalities etc.). However, given the burden of some of these conditions in certain populations, we do anticipate that these children will be included in the overall sample.
Type of exposure/intervention: We will include studies that compare the effect of the timing of introducing CF:
- Early introduction of CF (before 5 months of age) compared to 5 to 6.9 months of age
- Late introduction of CF (after 7 months) compared to 5 to 6.9 months of age
We will assess the effect of early and late introduction on the health, nutrition and developmental outcomes at any time point in life, regardless of the type of milk-feeding (breastfed, formula-fed or mixed-fed) provided to the infant. We will include studies that have appropriately defined the timing and constitution of CF (i.e. foods and beverages other than human milk or infant formula including liquids, semisolids, and solid food).
Types of Outcomes
Primary Outcomes
- Anthropometric Measures (Height/length; Height/length gain; Weight/Weight gain; Head circumference; Stunting (height-for-age z-score <-2 SD); Wasting (weight-for-height z-score <-2 SD); Underweight (weight-for-age z-score <-2 SD); Overweight; Obesity; BMI; HAZ; WAZ; WHZ)
- Morbidity (including fever, upper respiratory infections, lower respiratory infections, diarrhea, malaria, other infections, atopic dermatitis etc.)
- Anemia
- Child Developmental Measures (cognitive, socio-behavioral, motor etc.)
- Food preferences
- Dietary patterns/dietary diversity
- Food allergies
- Non-Communicable Diseases (NCD) including glucose intolerance, diabetes, hypertension, cardiac, inflammatory or auto-immune diseases, cancers
- Infant and child mortality
Secondary Outcomes
- Hemoglobin level
- Micronutrient status (including serum micronutrient levels for iron, serum ferritin, transferrin receptor, vitamin A, zinc, folic acid, B12 and fatty acids)
- Waist circumference
- Skinfold thickness
- Gut health and the microbiome
- Bone mineral density/bone mineral content
- Maternal outcomes: Lactational amenorrhea, maternal birth spacing
Search methods for identification of studies
Electronic search and search strategy : Using the key terms (Annex 1), we will search MEDLINE , EMBASE, Web of Science Index Medicus, CINAHL, Lilacs, CENTRAL (Cochrane Library) and eLENA (WHO), Index Medicus for the WHO Eastern Mediterranean Region (IMEMR), Western Pacific Region Index Medicus (WPRIM), Index Medicus for South-East Asia Region (IMSEAR), and African Index Medicus. We will also search for ongoing trials from clinicaltrials.gov, and non-indexed and grey literature from Google Scholar and key organizational websites. To identify any missing papers, we will search the bibliography of all included studies and all relevant systematic reviews. We will not employ any language or date restriction.
Searching other sources: We will search the reference list of all the included studies and relevant systematic reviews to look for studies missed during the electronic search. We will also put the title of included studies on google web and search the first 50 hits to identify any missing papers. We will also contact authors in case of missing/unpublished data.
Data collection and analysis
Selection of Studies: Within Covidence (online web-based systematic review tool), two review authors will independently screen for potential inclusion of all titles and abstracts identified as a result of the search (27). Following title and abstract screening, full texts will be independently screened for inclusion. Any disagreements will be resolved through discussion or by consulting a third review author, if required. Reasons of exclusion will be recorded for all the studies excluded at the stage of full text screening.
Data Extraction and Management: Data extraction for study characteristics and outcome data will be done in a data collection form. Two review authors will independently extract data and discrepancies will be resolved through discussion until consensus has been achieved or by consulting a third reviewer, if required. Attempts will be made to contact authors of included studies to obtain clarifications or additional data. We will extract data on the following study characteristics:
- Study Methods: journal, publication year, study design, total duration of study, study location, study setting, and withdrawals
- Participants: number, mean age, age range, gender, inclusion criteria, exclusion criteria infant feeding prior to and after introduction of CF, birth weight, gestational age, potential key confounders (for e.g. education, socioeconomic status, sex of caregiver, maternal/paternal age, race and/or ethnicity, milk feeding practices (breast milk, infant formula, or both))
- Interventions/exposures: intervention/exposure description, duration of intervention/exposure and comparison group description
- Outcomes: primary and secondary outcomes specified and collected, time points reported, and extraction methods used.
- Additional information: trial funding sources, study limitations and notable conflicts of interest of trial authors.
After data extraction, one review author will transfer data into Review Manager (RevMan) 5.4 software (28). We will double-check that data are entered correctly by comparing the data presented in the systematic review with the study reports. A second review author will spot check study characteristics for accuracy against the study reports.
Quality Assessment of Included Studies: Quality assessment will be done by using Cochrane risks of bias tool (29) for trials and risk of bias in non-randomized studies of intervention (ROBBINS-I) (30) for observational studies. Two independent authors will assess quality of all eligible studies and disagreements will be resolved by consensus or contacting a third author.
For trials, we will assess risk of bias according to the following domains:
- Random sequence generation
- Allocation concealment
- Blinding of participants and personnel
- Blinding of outcome assessment
- Incomplete outcome data
- Selective outcome reporting
- Other bias
We will judge each potential source of bias as high, low, or unclear and provide a quote from the study report together with a justification for our judgement.
For observational studies, the risk of bias will be judged based on confounding factors, selection of participants into the study, classification of interventions, deviations from intended intervention, missing data, measurement of outcomes, and selection of the reported results (30).
Measures of Treatment Effect: We will enter outcome data on RevMan software 5.4. Data from intervention and observational studies will be analyzed separately. Data will also be analyzed separately for the following two comparisons:
- Early introduction of CF (before 5 months of age) compared to 5 to 6.9 months of age
- Late introduction of CF (after 7 months) compared to 5 to 6.9 months of age
For dichotomous outcomes, we will use risk ratio (RR) while for the continuous outcomes, we will use mean difference (MD) or standardized mean difference (SMD) along with 95% confidence interval (CI). Data reported as a medians and interquartile ranges will be converted to means and standard deviations using standard formulas. We will undertake meta-analyses only where it is meaningful, i.e., if the exposures, participants, and the underlying clinical question are sufficiently similar for pooling.
Data Synthesis: Where data is available from two or more studies for a particular outcome, we will perform meta-analysis by pooling data on RevMan software. We will perform a random effects analysis for all comparisons, using inverse variance and Mantel-Haenszel methods to calculate the weights for continuous and categorical outcomes. This is a more conservative approach, as we expect the data to be heterogeneous.
Unit of Analysis Issues: We will conduct a meta-analysis separately for different study designs and for outcome subcategories. We will also analyze data of observational studies separately. For experimental studies, we will include both individually randomized trials and cluster-randomized in the analyses. For cluster-RCTs that have not been adequately adjusted for clustering, we will use the reported intra-cluster correlation coefficient (ICC) from trials’ original data sets along with the mean cluster size (M) to calculate the design effect. We will then use the methods set out in the Cochrane Handbook for Systematic Reviews of Interventions to calculate the adjusted sample sizes using the design effect (31). We will use an estimate of the ICC derived from the study (if possible), or from a similar study and study population if this is not possible. If we identify both cluster-RCTs and individually randomized trials that are similar in exposure and outcome assessment, then we will consider it reasonable to combine the results from both in one meta-analysis. We will meta-analyze effect sizes and standard errors using the generic inverse-variance method using RevMan software.
Assessment of Heterogeneity: Statistical heterogeneity will be assessed using τ2, I2, and significance of the χ2 test; we will also assess heterogeneity by visually inspecting forest plots. Based on prior clinical knowledge, we expect clinical and methodological heterogeneity in included studies and therefore, we will attempt to explain any observed statistical heterogeneity using subgroup analysis.
Assessment of Reporting Biases: For outcomes including more than 10 studies, we will create and examine a funnel plot to explore possible small-study and publication biases.
GRADE and Summary of Finding Tables
We will construct Summary of Finding (SoF) tables for primary outcomes for both the comparisons summarizing the quality of evidence according to the outcomes as per the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) criteria (32). It covers consideration of within‐study risk of bias, directness of evidence, heterogeneity, precision of effect estimates and risk of publication bias. We will rate the certainty of evidence for each key outcome as “high”, “moderate”, “low”, or “very low”. For non-randomized studies, the evidence quality will be upgraded based on large magnitude of effect, dose-response relationship, and effect of all plausible confounding factors in reducing the effect (where an effect is observed) or suggesting a spurious effect (when no effect is observed).
Sub-group Analysis
We will perform sub-group analysis for the following groups:
- By type of feeding before introduction of CF - exclusively breastfed, exclusively formula-fed, mixed breast and formula fed
- By timing of introduction of CF
- For the first comparison: <4 months, 4-5 months
- For the second comparison: 7-8 months, > 8months
- By age at outcome assessment – (till 23 months of age, 24-59 months, 5-10 years)
- By income regions - Low- and middle-income countries, high-income countries
- By birthweight - Normal birthweight, low birthweight, very low birthweight
- BY gestation at delivery – term, preterm
- By gestation – normal, small-for-gestational-age
Sensitivity Analysis
We will also perform a sensitivity analysis to examine only studies which had controlled for potential confounders. If numbers permit, sensitivity analyses will be performed on the primary outcomes to consider the impact of high risk of bias relating to sequence generation and/or allocation concealment.